텍스트 마이닝을 공부하기 위한 자료입니다. 언어에 상관없이 적용할 수 있는 자연어처리 / 머신러닝 관련 자료도 포함되지만, 한국어 분석을 위한 자료들도 포함됩니다.
- 이 자료는 현재 작업중이며, slide와 jupyter notebook example codes가 포함되어 있습니다.
- 이 자료는 soynlp package를 이용합니다. 한국어 분석을 위한 자연어처리 코드입니다. soynlp 역시 현재 작업중입니다.
- Slides 내용에 관련된 texts 는 blog 에 포스팅 중입니다.
- 실습코드는 코드 repository 에 있습니다.
- Python basic
- jupyter tutorial
- From text to vector (KoNLPy)
- n-gram
- from text to vector using KoNLPy
- Word extraction and tokenization (Korean)
- word extractor
- unsupervised tokenizer
- noun extractor
- dictionary based pos tagger
- Document classification
- Logistic Regression and Lasso regression
- SVM (linear, RBF)
- k-nearest neighbors classifier
- Feed-forward neural network
- Decision Tree
- Naive Bayes
- Sequential labeling
- Conditional Random Field
- Embedding for representation
- Word2Vec / Doc2Vec
- GloVe
- FastText (word embedding using subword)
- FastText (supervised word embedding)
- Sparse Coding
- Nonnegative Matrix Factorization (NMF) for topic modeling
- Embedding for vector visualization
- MDS, ISOMAP, Locally Linear Embedding, PCA, Kernel PCA
- t-SNE
- t-SNE (detailed)
- Keyword / Related words analysis
- co-occurrence based keyword / related word analysis
- Document clustering
- k-means is good for document clustering
- DBSCAN, hierarchical, GMM, BGMM are not appropriate for document clustering
- Finding similar documents (neighbor search)
- Random Projection
- Locality Sensitive Hashing
- Inverted Index
- Graph similarity and ranking (centrality)
- SimRank & Random Walk with Restart
- PageRank, HITS, WordRank, TextRank
- kr-wordrank keyword extraction
- String similarity
- Levenshtein / Cosine / Jaccard distance
- Convolutional Neural Network (CNN)
- Introduction of CNN
- Word-level CNN for sentence classification (Yoon Kim)
- Character-level CNN (LeCun)
- BOW-CNN
- Recurrent Neural Network (RNN)
- Introduction of RNN
- LSTM, GRU
- Deep RNN & ELMo
- Sequence to sequence & seq2seq with attention
- Skip-thought vector
- Attention mechanism for sentence classification
- Hierarchical Attention Network (HAN) for document classification
- Transformer & BERT
- Applications
- soyspacing: heuristic Korean space correction
- crf-based Korean soace correction
- HMM & CRF-based part-of-speech tagger (morphological analyzer)
- semantic movie search using IMDB
- TBD
자료를 리뷰하고 함께 토론해주는 고마운 동료들이 많습니다. 특히 많은 시간과 정성을 들여 도와주는 태욱에게 고마움을 표합니다.